Open AccessApproaches to stream solute load estimation for solutes with varying dynamics from five diverse small watersheds
Author(s) -
Aulenbach Brent T.,
Burns Douglas A.,
Shanley James B.,
Yanai Ruth D.,
Bae Kikang,
Wild Adam D.,
Yang Yang,
Yi Dong
Publication year - 2016
Publication title -
ecosphere
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.255
H-Index - 57
ISSN - 2150-8925
DOI - 10.1002/ecs2.1298
Subject(s) - watershed , environmental science , sampling (signal processing) , hydrology (agriculture) , water quality , snowmelt , biogeochemical cycle , dissolved organic carbon , atmospheric sciences , ecology , surface runoff , environmental chemistry , chemistry , geology , geotechnical engineering , filter (signal processing) , machine learning , computer science , computer vision , biology
Estimating streamwater solute loads is a central objective of many water‐quality monitoring and research studies, as loads are used to compare with atmospheric inputs, to infer biogeochemical processes, and to assess whether water quality is improving or degrading. In this study, we evaluate loads and associated errors to determine the best load estimation technique among three methods (a period‐weighted approach, the regression‐model method, and the composite method) based on a solute's concentration dynamics and sampling frequency. We evaluated a broad range of varying concentration dynamics with stream flow and season using four dissolved solutes (sulfate, silica, nitrate, and dissolved organic carbon) at five diverse small watersheds (Sleepers River Research Watershed, VT ; Hubbard Brook Experimental Forest, NH ; Biscuit Brook Watershed, NY ; Panola Mountain Research Watershed, GA ; and Río Mameyes Watershed, PR ) with fairly high‐frequency sampling during a 10‐ to 11‐yr period. Data sets with three different sampling frequencies were derived from the full data set at each site (weekly plus storm/snowmelt events, weekly, and monthly) and errors in loads were assessed for the study period, annually, and monthly. For solutes that had a moderate to strong concentration–discharge relation, the composite method performed best, unless the autocorrelation of the model residuals was <0.2, in which case the regression‐model method was most appropriate. For solutes that had a nonexistent or weak concentration–discharge relation (model R 2 < about 0.3), the period‐weighted approach was most appropriate. The lowest errors in loads were achieved for solutes with the strongest concentration–discharge relations. Sample and regression model diagnostics could be used to approximate overall accuracies and annual precisions. For the period‐weighed approach, errors were lower when the variance in concentrations was lower, the degree of autocorrelation in the concentrations was higher, and sampling frequency was higher. The period‐weighted approach was most sensitive to sampling frequency. For the regression‐model and composite methods, errors were lower when the variance in model residuals was lower. For the composite method, errors were lower when the autocorrelation in the residuals was higher. Guidelines to determine the best load estimation method based on solute concentration–discharge dynamics and diagnostics are presented, and should be applicable to other studies.